Urban Technology at University of Michigan week 241

You Can't Spell Architecture Without AI - an Interview with Eric Cesal

While the name urban technology implies something that includes buildings, a lot of our time is focused on the larger scales of neighborhoods and cities at large. Lately we’ve had some occasion to think about architecture specifically, zooming into questions of buildings and data, and how more data about buildings is (maybe!) reshaping architectural practice. In that spirit, this week we are bringing you an interview with Eric Cesal, who is an architect with a unique voice at the intersection of architecture and AI.

When generative AI that can produce images such as Dall-E and Midjourney launched, such tools were picked up almost immediately by architects and urban planners. Planning education tends not to include a strong focus on renderings and image-making, so what we observed in those circles was a wide spectrum of bemusement. Sentiments such as, “hey look, I can make pictures of streets that have been turned back into parks” from planners dotted my timeline. On the architecture side there was a noticeably more equivocal mood. “We’re the makers of urban imagery. Will these algos steal our jobs?” mixed with “whoa” and explorations to the effect of, “can this make my work better?” But of course, making images is only part of the architect’s job, and it’s the everything else that has occupied much of Eric’s thinking as well as our conversation below.

---

💬 Hello! This is the newsletter of the Urban Technology program at University of Michigan, in which we explore the ways that data, connectivity, computation, and automation can be harnessed to nurture and improve urban life. If you’re new here, try this short video of current students describing urban technology in their own words or this 90 second explainer video.

---

🤖 You Can’t Spell Architecture Without AI

Eric J. Cesal is a multifaceted professional whose career blends design, education, and humanitarian efforts. Known for his leadership in post-disaster reconstruction, Eric has been instrumental in rebuilding communities following major catastrophes such as the Haiti earthquake and Superstorm Sandy. Eric’s book, “Down Detour Road: An Architect in Search of Practice,” serves as both a memoir and a manifesto, offering insights into the architectural profession and its societal responsibilities. ←That part of the introduction was written by GPT. This part is written by Bryan → These days Eric is also a prodigious author here on substack, where his publication Life as a Disaster is largely dedicated to an exploration of AI in the built environment. It’s a rare place that blends sharp thinking and terminator meme pics.

BRYAN BOYER: When I came across your video “Everyone-An-Architect-Pt1” [article link / video link] it was the first time that I encountered material exploring AI and architecture that focused on the process rather than fixating on an image of some kind. What I appreciate about this is that it provides a counter balance to the “prestige” technologies that often get too much attention within architectural discourse. What’s different about your approach?

ERIC CESAL: I have been examining ways in which AI will infiltrate architectural practice that we don't necessarily initiate…. That aren't necessarily good for us [as architects] and that we may not actually see coming. Those are the areas that very much interest me.

Going back to November of 2022 when chat GPT debuted, it was just a chat bot. But, to crib from Yuval Harari, language is the thing that all of our other things are made of. Once you can do language, how far can you get in approximating the work of an architect?

We don't [often] think about it in those terms because architectural work is considered visual and spatial. But there is quite a bit of that visual and spatial work that's built on top of language and linguistic work. So that's where I began.

BRYAN: In your predictions essay ‘Design in 2025: the near future of architecture & AI” you wrote about clients getting more capable thanks to their use of AI tools to do things like generate images of their desired spaces. Should architects stop this use of AI or roll with it?

ERIC: More so the latter, but only because I don't think there would be a way to stop it from happening. The idea itself probably has two sources. One is from friends in the medical field who complain constantly about patients that'll come in and say, “Hey, I Googled my symptoms and I have Ebola.” A doctor’s first conversation is now just walking the patient back from whatever conclusions they've come up with on their own. The second source was the early conversation in the architectural field around AI-centered image generators such as DALL-E and Midjourney.

Designers took to these tools like water and loved coming up with imagery, but I thought it was a curious omission that nobody ever seemed to talk about the fact that everybody else had access to these tools as well.

Visualization has been one of the skills that architects and other designers have, and we take it for granted that visualization is this superpower that most people do not have. When an ordinary person has an idea, they don't have any way to represent it visually in two dimensions or three dimensions. Now, thanks to these tools, they do.

So between those two things, I started thinking about how that puts an evolutionary pressure on the architect-client relationship and the logical conclusion is that clients will come in with something from Midjourney or another program and say, “This is what I want.” If architects capitulate to that, they're neglecting their professional duties and professional identity.

But I don't think it works to turn away or turn our noses up at it, either, because image generation is going to be a ubiquitous technology. We have to develop a new kind of vocabulary to explain that image making is part of practice but not the whole of it. How you walk a client back from some generated images that look good but may not be viable depends on the architect and the client. We are going to be having that conversation in the future, and we should start practicing.

BRYAN: What are your habits for cutting through AI-related hype?

ERIC: I try to get most of my information from academic sources, and that's something I recommend to all audiences. I like to read academic papers and draw my own conclusions because, I do feel that the academic stuff is more rigorous, so I go there first. From there, I try to root my hypothesizing in some first principles.

For instance, by asking myself whether a particular technological development has any reason to exist… Are there people who would be pushing it? Is there a market demand for whatever is being proposed? And fundamentally: is what is being proposed superior to what we're doing now? I try to ground it in the most fundamental things I can, which is academic research first principles, market dynamics, human psychology, social psychology… these sorts of things.

It's difficult; there's so much hype out there. Did you see my piece on AI as a smart horse? The basic premise was that conventional economics divides everything into two classes, right? Materials and labor. And AI isn't really either of them. It doesn't quite fit in either one of those two categories. Instead it's like a draft horse. AI is going to work for us. It's going to do the things that we can't do. We're going to train it, and supply it with water for cooling, and energy for running and this sort of thing.

BRYAN: How do you see the evolution of the smart home?

ERIC: We have conflated the term “smart building” with, basically, thermostats, right? Or maybe very sophisticated lighting and HVAC controls. What would happen if a building had a brain? What would it do? What could it do?

I've had this idea about AI-enabled buildings—sentient buildings if we can call them that—as a running design partner. So, if I'm an architect and design all these buildings with embedded AI intelligence, at some point the buildings can collectively be a thoughtful partner in the design of my next project. I can sit down to design something and ask all the different AIs, “Hey, what's working inside your spaces?”

BRYAN: What type of sensor infrastructure do you think needs to happen for something like that to have a baseline of utility?

ERIC: We have to jump over a massive privacy chasm to start answering that question. You can fill the whole thing up with sensors t but nobody would want to work or live there because you're under constant surveillance.

The other possibility is how much data can that building pick up as a result of the AI that's already embedded in other things. If we're all walking around with our phones and the AI can get access to that, then it understands which areas of the building are being used and which aren't without having any sort of dedicated sensors. Robots are another one. Once robots become relatively ubiquitous, they are another data gathering channel.

IoT has been a heavy lift for the last 10 years because of the cost of installing, maintaining, and programming these dedicated sensor networks, and now it's very likely that the intrusion of AI into devices like computers, laptops, phones, and watches,, as well as the proliferation of robots, may give us the sensor networks we're looking for without having to have dedicated sensors in place.

With those already-present devices acting as sensors, the privacy issue is real but you have to figure there's a way to collect data in an anonymous way that people will accept. The amount of data that my phone collects on me is extreme! If you could take me back in time 20 years and ask me to sign a contract that says, “Hey, will you carry a device that monitors your every movement and everything you say?” My response would be: No! But we just get adjusted to the fact that data is being collected on us. I'm not sure if we want to accelerate or continue that trend, but it's a reality. People are worried about AI invading your privacy, but your phone has already done that.

BRYAN: Really what we're talking about is how decision-making changes when available information changes. How does the work of the architect change when all buildings create data exhaust streams?

ERIC: One of the more dysfunctional aspects of architecture is the endurance of this genius hero myth, where design decisions are made after somebody gets struck by a lightning bolt.

Architects need to cultivate a position that is elevated, respected, and trusted. But the way to achieve this is to do the work] out in the open and in plain sight, like a good leader.

Fundamentally, the architecture process today celebrates the opaque. But if data becomes more prevalent in design decision-making, will digital design firms start to perform as financial firms do, with a stream of constant data informing how they're making their decisions?

I think that necessarily pushes those decisions to become more transparent and democratic. We can all sit around the design, look at the data, and draw conclusions. We don't have to put our faith and trust in one person in the corner of the office whose judgment is both opaque and irreproachable.

The point is, we can't have a value proposition in public that's based on mystery. We can't ask the public to “trust us” in an age of AI.

BRYAN: What’s your favorite city and why?

ERIC: If I had to choose one I’d probably say Kathmandu. It’s the nexus of all we built. Old, new, East, West, Up, down. If there’s a Grand Central Station for Human Civilization, it’s probably Kathmandu.

---

These weeks: Cities Intensive preliminary overview with students. Spring break. UT faculty committee meeting. Graduation event planning, woah! The birds are chirping, and loud. 🏃